On Tue, Aug 27, 2019 at 10:44:39AM -0700, Omar Sandoval wrote:
> On Tue, Aug 27, 2019 at 02:35:13PM +0200, David Sterba wrote:
> > On Thu, Aug 15, 2019 at 02:04:03PM -0700, Omar Sandoval wrote:
> > > From: Omar Sandoval <osan...@fb.com>
> > >
> > > In btrfs_file_write_iter(), we treat a write as synchrononous if the
> > > file is marked as synchronous. However, with pwritev2(), a write with
> > > RWF_SYNC or RWF_DSYNC is also synchronous even if the file isn't by
> > > default. Make sure we bump the sync_writers counter in that case, too,
> > > so that we'll do the CRCs synchronously.
> > >
> > > Signed-off-by: Omar Sandoval <osan...@fb.com>
> > > ---
> > > fs/btrfs/file.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
> > > index 4393b6b24e02..27223753da7b 100644
> > > --- a/fs/btrfs/file.c
> > > +++ b/fs/btrfs/file.c
> > > @@ -1882,7 +1882,7 @@ static ssize_t btrfs_file_write_iter(struct kiocb
> > > *iocb,
> > > u64 start_pos;
> > > u64 end_pos;
> > > ssize_t num_written = 0;
> > > - bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
> > > + bool sync = iocb->ki_flags & IOCB_DSYNC;
> >
> > I'd like to merge the patches 1-3, but have hard time matching the
> > changelog to the change here. It's from one set of sync flags to
> > another, mentioning pwritev2 but that's a syscall and the function
> > itself does not use the sync flags at all. That's probably somewhere
> > deep in the vfs calls but that's what I'd appreciate stated explicitly
> > in the changelog as I was not able to find it out in a reasonable time.
>
> You're right, there are a few layers here. How about this for the
> changelog:
>
>
> The VFS indicates a synchronous write to ->write_iter() via
> iocb->ki_flags. The IOCB_{,D}SYNC flags may be set based on the file
> (see iocb_flags()) or the RWF_* flags passed to a syscall like
> pwritev2() (see kiocb_set_rw_flags()). However, in
> btrfs_file_write_iter(), we're checking if a write is synchronous based
> only on the file; we use this to decide when to bump the sync_writers
> counter and thus do CRCs synchronously. Make sure we do this for all
> synchronous writes as determined by the VFS.
That's great, thanks, no need to resend.
